21th International Conference on Production Research

A NEW ALGORITHM FOR THE DESIGN OF A TRANSPORTATION FLEET

R. Pea, A. Moran, J. Seplveda
Industrial Engineering Department, University of Santiago of Chile, 3769 Ecuador Ave. P.O Box 10233,
Santiago, Chile


Abstract
In this work, a new strategy and algorithm for dimensioning transportation fleets for the delivery of consumer
products from a distribution center (DC) to retailers is presented. The algorithm named CK, works on a
horizon in which a delivery forecast is deployed. Given a set of vehicles and the delivery orders forecast, a
volumetric optimization routine determines the optimal length of a container having a standard width and
height. A set of parameters such as the desired customer service level, concentration level (i.e., expected
deliveries per day), and expected growth rate, give the necessary flexibility for the decision making process.
The CK algorithm simulates the behavior of costs for a finite number of test fleets and chooses the most
efficient. Efficiency is measured as the best assignment of volume to a delivery in a given transport. The CK
approach was implemented in a manufacturer and distributor of appliances in Santiago of Chile leading to a
decrease of 25,8% of the annual spending in transportation services, an increase of 35% in utilization of the
available capacity, and an increase from a historical performance of 95% in customer service level to a
99,93%.

Keywords:
Fleet sizing; logistics; supply chain; simulation.


1 INTRODUCTION
Modern organizations are not only concerned with
manufacturing a final product or producing a service but
they also intend to provide a high customer service. In this
regard, the transportation strategy is particularly important
since an efficient movement of goods contributes to
increased market competitiveness, greater economies of
scale in production, and lower prices along the supply
chain [1].
Inefficiency in a transportation system is mainly due to the
lack of knowledge about:
1. The distribution of the volume historically transported,
since the irregularity of the demand volume of
deliveries is unknown and therefore the vehicle
capacity in the fleet would not have to be uniform.
2. The suboptimal placement of products in the vehicle
loading space, since even though the irregularity of
the dispatched volume is recognized, volumetric
optimization of vehicles in the fleet is not applied
3. The concentration of dispatch orders during the
working period, since if during a period the deliveries
cannot be concentrated or consolidated, an adequate
daily transportation scheduling will not be achieved.
A strategy for selecting a fleet must necessarily integrate
the aspects mentioned above [2][3]. Figure 1, shows the
efficiency of a fleet in terms of transportation costs, based
on the size of the vehicle. A well designed fleet should be
well behaved for a given planning horizon and badly
behaved outside that horizon. This behavior is
consequence of an inadequate fleet management. The
fleet historically designed (dashed line in Figure 1) ignores
the distribution of the transported volumes in the past
period, and therefore the vehicle capacity has no
relationship with the deliveries in the present period; it also
misses the volumetric optimization so that vehicle capacity
is not well utilized, and vehicle assignment in the daily
schedule is poorly made.













Figure 1: Efficiency of a transportation fleet

The contribution of this paper is for improving the
managerial practices in the area of distribution logistics.
For this purpose, it is developed a strategy which consists
of five elements, as follows:
1. Data inputs: dispatched invoices in former periods and
product dimensions for generating a forecast of the
volume to be transported.
2. Control Variables: expected service level of transport,
concentration level of deliveries, and expected growth
rate of the volume.
3. CK algorithm: generator of a finite number of test fleets
for which their behavior in transportation costs is
simulated.
4. Planner: agent which plans the strategy by entering
information inputs and handling the control variables.
5. Information Output: the efficient transportation fleet
determined by the CKs algorithm, which satisfies the
planners requirements.

Figure 2: CK Strategy

The strategy have been implemented in the Distribution
Center (DC) of a Chilean company named Sindelen S.A.,
which manufactures and distributes different appliances
such as, stoves, refrigerators, gas heaters, hair dryers,
toasters, etc. The customers are the main retail chains of the
country, as well as independent stores in several cities. The
fleet obtained with the application of the CK strategy helped
to reduced the annual transportation cost in 25,8%
(monetary efficiency), increased in 35% the utilization of the
transportation capacity (process efficiency), and rose the
service level up to 99,93%, as compared with the 95% of the
historically designed fleet (service improvement).

2 CK STRATEGY
In order to generate the CK algorithm, some assumptions for
the global strategy must be established.
2.1 Assumptions
The transportation cost rate depends on the vehicle
containers capacity; i.e., the greater the capacity is, the
greater is the opportunity cost for the transporter and
therefore the price for the dispatcher.
The transport capacity is determined by the length of
the container, since the front surface practically remains
constant for a wide range of lengths (5,7 m
2
in average)
The regularity of the demand for products allows the
volume demand to be also regular, since the product
size remains unchanged between periods.
2.2 CK General Procedure
Figure 2 depicts the general procedure.
1. The human planner defines the planning horizon on
which the database will be generated.
2. The planner defines the set of vehicles from which the
transportation fleet will be selected.
3. The planner enters all the dispatch orders in the
previously defined horizon, as in step 1.
4. A software agent for volumetric optimization (VO)
calculates the required volume for the product list in each
dispatch order.
5. The VO software agent determines the optimal length
with a standard front dimension.
6. The optimal length (OL) is recorded in the database with
additional data about the dispatch.
7. The planner determines the input level of the control
variables; i.e., the expected service level, concentration
level, and expected growth.
8. The CKs algorithm generates a finite number of test
fleets and simulates the transportation costs. Then, the
least cost fleet is selected.

3 CK ALGORITHM
The construction begins with the definition of the planning
horizon on which the forecasted volumes will be generated.
In what follows, this section explains the notation, concepts,
and relationships utilized by the CK strategy and the logic of
its algorithm.

Planning Horizon: Let } ,..., , { =
2 1 H
s s s S be the set of sub-
periods and } ,..., , { =
2 1 D
i i i I be the index set of days per
each subperiod, then the length of the planning horizon and
the number of days per subperiod are defined as:
S H = (1)
I D =

(2)
Where x stands for the cardinality of set x .
The planning horizon will allow to collect information
regarding the volume that has been transported during the
period H . Then, by using a growing factor or multiplier, the
fleet will be forecasted for the next year over the same
defined horizon. It is recommended that H be annual and
the values of S be monthly. The next step is to define a set
C of existing vehicle types available from providers in the
market, which will be classified by the planner according to
their capacities.

Fleet: Let } ,..., , { =
2 1 N
C C C C be the set of N classes of
transporters, each with capacity
j
C ; their transportation
21th International Conference on Production Research
costs are denoted by } ,..., , { =
2 1 N
C C C
Y Y Y Y
.
j
C
K is the
quantity K of vehicles per each class
j
C . Thus the p-th
fleet is expressed as:
} ,..., , { =
2 1 N
C C C p
K K K X (3)
Table 1 shows the set defined by the planner for the case
study described in this paper.
Table 1: Set C
Intervals
j
C
Y
j
C
j
C
L

Front Capacity
(m) CLP (m) (m
2
) (m
3
)
[0 - 4] 10,500 A 4 5.72 22.88
(4 - 5] 17,330 B 5 5.72 28.60
(5 - 7] 25,200 C 7 5.72 40.04
(7 - 8] 55,000 D 8 5.72 45.76
(8 - 13] 80,000 E 13 5.72 74.36

The planner has available vehicles of different capacities
and therefore, he/she has different transportation costs to
deal with. It is assumed that the cost remains constant in a
small capacity range. In Table 1,
j
C represent a transport
class with a cost of value
j
C
Y and capacity
j
C
L . For
instance, from Table 1 a fleet with the vector } 5 , 4 , 8 , 7 , 1 { = X
means that types C-D-E require 8, 4, and 5 trucks with
capacity over 40 cubic meters, respectively.

Fleet size: The size M of the p-th fleet is given by:

=
=
N
j
C p
j
K M
1
(4)
For instance, 25 5 4 8 7 1 = + + + + = M vehicles for the vector
above. Once the planning horizon and the set C are
defined, the fleet design process consists of finding the
capacities and quantities to be selected. That is, the values
) , ( K C for which the transportation costs are minimized.
Efficiency in this case is measured by the best volume
assignment of a given dispatch to a vehicle in the set C.

As shown in Figure 2 before data is recorded onto the
database, the VO algorithm determines 1) the optimal
volume for the dispatch and 2) the optimal length OL to
contain the load. Given the value of OL of a dispatch Q in a
day i in the period s, it is possible to assign an optimal
vehicle
j
C of length
j
C
L
Efficient capacity: Let
s i
Q
OL
,
be the optimal length of a
vehicle of standard height and width able to efficiently serve
the dispatch Q, in the day i of period s; then, there exists a
vehicle
j
C , of length
j
C
L for which the condition (5) holds.
j
C
s i
Q
L OL
,
(5)
The determination of an efficient capacity is a process for
assigning optimal volumes in order to minimize
transportation costs. After the efficient capacity has been
assigned to each group
j
C , the number of vehicles must be
counted for determining the total length required as in a
infinite capacity planning.
Upper bound for fleet size: Let
s i
Q
L
,
be the total length of
efficient vehicles of type
j
C assigned in the day i of sub-
period s; then an upper bound is defined by (6).
j
J
j
C
s i
C
s i
C
L
L
K

=
,
,
(6)
Formula (6) states that for day of subperiod ,
s i
C
j
K
,
vehicles of length
s i
C
J
L
,
are needed, or a relation one to
one between vehicles and dispatches. This upper bound
means a fleet of a size equivalent to the sum of all
dispatches to be delivered during the planning horizon H
(i.e., Q M =

with M a big number). Thus, control variables
must be used to adjust the size to the actual transportation
needs. One of these control variables is a parameter that
measures the number of dispatches a vehicle is able to
make in a given day. This is determined by the agenda
manager according to the time windows that have been
agreed with the customer [4]. That is, the agenda manager
is defined as the agent able to generate appointments
between the supplier and the customer to sent and receive
the shipments, respectively. During a day, a vehicle can
make more than one dispatch or appointment. The more
appointments the vehicle can make the lesser is the
concentration level. The planner seeks to minimize
concentration but such a decision is influenced by other
external policies, such as:
Stock Policy: since a poor policy with frequent stockout
forces the manager to postpone a dispatch until the
stock and the receiver are available.
Customer Policy: since the receiver may force the
agenda manager to set the time of a delivery, or else to
delay the service.
Despite of the conditions that influence an agenda, it is
possible for the planner to establish a concentration level as
defined below.
Concentration level: Let R be the maximum number of
shipments a vehicle is able to make in a day; then the
concentration level is defined by (7).
R
1
=

R , (7)
In a given working shift, there will be a maximum of R times
to make shipments which concentrate the % of total daily
orders. The importance of this definition is that is a
control variable depending on the planners judgment and
the stock and customer policy, as mentioned above.
For each level, fleets with different values in transportation
costs will be generated. Hence, the least transportation cost
is not a sufficient condition for determining an efficient fleet
and therefore the expected transportation service level
needs to be introduced as follows.
Expected transportation service level:. Let
p
X be the p-
th fleet generated with a concentration level . Then, the
planner is able to define the behavior of the p-th fleet by (8).
Dispatches Total
Dispatches Unfilled
1 = (8)
As the planner allows a number of unfulfilled dispatches in a
test fleet ( % 100 ), a penalty cost much greater than the
cost of transporting must be assigned [5], as follows.
Penalty for unfulfilled dispatches: Let be the expected
transportation service level with % 100 = ; then there exists a
cost, say
)% - 1 (
Y with )% - 1 ( of unfulfilled dispatches in the
planning horizon H . The penalty is given by (9).
)% - 1 (
= Y Y (9)
The determination of an efficient fleet requires a database of
the historical volume that has been transported (see Figure
2). This information will allow the planner to forecast the next
year, according to the sales plan; that is, a growth rate can
be estimated for each period S s .The expected growth
level of the volume will be denoted by .

Expected growth level of volume: Let
s
V be the total
volume transported in subperiod S s , and
' s
V be the total
volume to be transported during the subperiod s of the next
year; thus, the growth factor for each subperiod S s is
given by (10).
' s
s
s
V
V
= (10)
Now, it is possible to concentrate daily dispatches by
utilizing the concentration factor

and in this manner to
assign a number R of shipments per vehicle. The
transportation requirement

B is defined next.

Transportation Requirement: Let } 0 {
0
= , and
s i
C
j
W
,

be the requirement for transportation of type
j
C in the day i
of subperiod s and
s
the expected growth factor of the
volume in the subperiod s ; then, for a concentration level
of dispatches:
s i
C s r
s i
C
j j
K W
, ,
) + 1 ( = (11)
Thus, the requirement
s i
C
j
B
,
is calculated by (12):


s i
C
j
W
,
; if
0
,

s i
C
j
W
=
s i
C
j
B
,
(12)

1 ] [
,
+
s i
C
j
W E ; if
0
,

s i
C
j
W

Where [ ] x E corresponds to the integer part function of x .It
is important to verify the condition for and
s
, since the
infinite fleet
s i
C
j
K
,
breaks down the dispatches in , passing
from a relation one to one of vehicles shipments to an one
to R relation, so that vehicles ship R

times during the daily
working shift [4].
A factor 0 =
s
helps to design a fleet for a given planning
horizon which allows to compare its efficiency with a fleet
that operated already during that horizon, whereas a value
0
s
will project from the database a new fleet for the next
years planned horizon. The quantity in (12) may have a
unique behavior throughout the periods since it involves
volume requirements depending on the demands regularity.
Then, a parameter must be defined in order to decide the
quantity B

of vehicles that is most representative for all the
subperiods. The parameter may follow any rule such as:
mode, mean, median, maximum, minimum, or mixed rules.
Even though this parameter changes the fleet composition, it
is not considered as a control variable since it does not
depend on the planner but on a rule that best fits to the
distribution of the transportation requirements.

Test quantity: Let } ..., , , { =
3 2 1 T


be a set of
T criteria defined by the planner; then the quantity of test
vehicles for the group
j
C is given by (13), with
s i
C
j
B
,
as
defined above.
=
,
,
U
s i
s i
C T C
j j
B K (13)
It is important to remark that the quantity in (14) is for testing
since by using the CKs algorithm, by simulation different
fleets will be generated with various concentration levels ,
satisfying an expected transportation service level , with a
projection
s
, and best efficiency criteria
T
. This is
formalized with the definition below.

Test fleet: Let be a set of criteria defined by the planner;
then the p-th test fleet is defined by (14).
} ,..., , , { = ) ; , (
3 2 1 N
C C C C p
K K K K X (14)
Which has associated cost ) ; , (
P
Y
.

For determining the efficient quantity * K different scenarios
must be evaluated, where the planners inputs are tested
one by one on the test fleets generated, as in (14). This
process is made by two types of assignments: progressive
and regressive, which are obtained only by simulation. Once
a test fleet ) ; , (
p
X has been defined, its availability
during a working shift must be determined as described
next.

Available fleet: With
j
C
K as the quantity of vehicles of
type
j
C , and } ,..., , { =
2 1 R


the set of R concentration
levels as defined by the planner, the quantity of available
trips of transport type
j
C is
j
C
AF where:
] [ = ] [ =
1 -
J j j
C C R C
K R E K E AF (15)
Even though there may be available trips in the assignment
process, it is possible that the transportation needs are not
covered. Then the progressive assignment is defined as
follows.
Progressive assignment:. The remaining quantity, or
residual quantity of transport,
s , i
C
j
RQ , and the progressive
assignment
s i
C
j
PA
,
are determined in two steps:
Step 1: 1 = j



j
j
C
s i
C
VD K
,
; if 0 AF K
j j
C
s , i
C
-
=
s , i
C
j
RQ (16)

0 ; if 0 AF K
j j
C
s , i
C
< -
21th International Conference on Production Research



s i
C
j
K
,

; if 0 K AF
s , i
C C
j j
-
=
,s i
C
j
PA (17)

j
C
AF ; if 0 K AF
s , i
C C
j j
< -
Step 2: 1 > j

s , i
C
s , i
C C
s , i
C
1 j j j j
RQ K AF - - = (18)

0 ; if 0
,

s i
C
j

=
s , i
C
j
RQ (19)

s i
C
j
,
; if 0
,
<
s i
C
j




j
C
AF

; if 0 RQ
s , i
C
j
>
=
,s i
C
j
PA (20)

s , i
C
s , i
C
1 j j
RQ K - ; if 0 RQ
s , i
C
j

In the assignment, the term progressive refers to the fact
that the transportation needs are covered from the smallest
fleet ( 1 = j ) to fleets with greater capacity ( 1 > j ) in a
progressive manner. If in the working shift for a type of
transport
j
C , the available trips
j
C
AF are insufficient to
cover the needs 0 RQ
s , i
C
j
then the fleet with highest
capacity in progressive order will have to cover it. Finally, if
for the type of transport with the highest capacity
N
C there
still are needs uncovered, then the dispatch must be
partitioned in vehicles of smaller capacity but with available
trips.

Regressive assignment: if the number of vehicles of type
j
C with free trips is
s , i
C
j
FT , then:
s , i
C C
s , i
C
j j j
PA AF FT - = (21)
In order to break down the dispatch of a vehicle of type
j
C
into smaller vehicles, the free length is defined as follows.
Free length:. With
s , i
C
j
FT as the total free trips of vehicles of
type
j
C and } ,..., , { =
2 1 N
C C C C
L L L L

as the lengths of
vehicles in

C, then there may be a length that has not been
utilized, as defined by (22).
j j j
C
s , i
C
s , i
C
L FT FL = (22)
Thus, the number of vehicles that are needed
is - =
, , s i
C
s i
C
j j
RQ

Length shortage: Let
N
C
L be the length of vehicles of type
N
C , then the length shortage
s , i
C
N
LS is represented by (23).
N N N
C
s , i
C
s , i
C
L * RQ LS = (23)
Then, the regressive assignment needs further definitions as
given by (24) (25) (26).
s , i
C
s , i
C
s , i
C
j N j
FL LS - = ; 1 = j
(24)
s , i
C
s , i
C
s , i
C
1 j N j
FL - = ; 1 j (25)

s , i
C
j
FT ; if 0
,
>
s i
C
j

=
s , i
C
j
RA
-
, ,
,
j
N j
j
C
s i
C
s i
C
s i
C
L
LS FL
E FT ; if 0
,
=
s i
C
j
(26)
0 ; if 0
,
<
s i
C
j

The final assignment
s i
C
j
FA
,
in the working shift is:
s , i
C
s , i
C
s , i
C
j j j
RA PA FA + = (27)
In the last definition (27) there still may be shortage of
transport capacity since the test fleet does not assure total
fulfillment, then the number of vehicle shortage is stated by
equation (28).

0 ;if 0
,

s i
C
N

=
s , i
C
N
VS

N
N
C
s i
C
L
E
,

;if 0
,
>
s i
C
N
and

N
N
C
s i
C
L
E
,

(28)

1
,
+

N
N
C
s i
C
L
E

;if 0
,
>
s i
C
N
and

N
N
C
s i
C
L
E
,

Finally, for finding the pair *) *, ( K C the costs are calculated
for each test fleet, for each of the periods s , with the control
variables defined by the planner, as follows:
+ = ) ; , (

, ,
i
s i
C
i
C
j
s i
C T R s
N j j
VS Y Y FA Y

(29)
The cost over the planning horizon H

of the proponed fleet
is given by (30).

) , ( = ) , , (
;
S
T R s T R H
Y Y (30)
The last step in the simulation is the finding of the fleets with
the least cost for a given expected service level . The
expression (31) summarizes the process.

- =
i S
s , i
C
i S
s , i
C
j
N
AF
CF
1 , then for :


} , , )}, , , ( min{ = , , { = t r s Y Y j i FA X
t r s s
i
C S
j
(31)
The efficient fleet is a vector } ,..., , { =
2 1 N
C C C
K K K v , in terms
of cost, a set gathering the best test fleets.
U
S
S H
X X = (32)


Formula (33) gives the fleet with the least annual cost.
} , )}, , , ( min{ = , , { = t r Y Y j s FA X
t r H H
s
C H
j
(33)

4 CK TESTING
The CKs algorithm was tested with data of the distribution
center of Sindelen S.A. It was implemented in an Excel
spreadsheet and the volumetric optimization software was a
free version of CubeMaster .In Table 2, the data entered
by the planner is shown.

Table 2: Input Data
Input Valor
H 2009
S } 12 ,..., 2 , 1 { (months)
I } 31 ,..., 2 , 1 { (days)
N 5 (transport classes)
C } , , , , { E D C B A
j
C
Y } 000 , 80 ; 000 , 55 ; 200 , 25 ; 300 , 17 ; 500 , 10 { ($CLP)
Y
000 , 100 ($CLP)
Q } 2712 ,..., 2 , 1 { dispatches
s i
q
LO
,

2712 (optimal lengths)
% 95

} 5 . 0 {

{ ; 006 . 0 ; 297 . 0 ; 155 . 0 ; 283 . 0 ; 080 . 0 ; 176 . 0 ; 137 . 0
359 . 0 ; 233 . 0 ; 230 . 0 ; 287 . 0 ; 181 . 0 }

{mean, mode, median, max, min, mix* }

The planner first evaluates the base period in 2009
( 0 = and he/she sets the control variables and ; only
one efficient fleet is generated for each period . S

Table 3 shows the generated fleets for the best selection
criteria

for the year 2009. This analysis by subperiod
determined an irregular fleet with an annual transportation
cost of CL$ 69,540,600 or USD 140,000. The transportation
service level is 99,51% in average with a mean of 17
vehicles. However, by performing an annual analysis with
variation of in 0.5 and 1, an efficient fleet for the planned
with a greater service level and reduced fleet size M was
determined. Table 4 shows the results of the analysis for a
fleet with 5 . 0 = . That is, by concentrating twice a day the
50% of the daily dispatches. Finally, it is possible to
introduce a value in the growth variable thus generating
an efficient fleet for the year 2010. The obtained fleet is
} 2 , 2 , 2 , 2 , 5 { =
2010
X also with 5 . 0 = . This fleet, as in 2009,
concentrates vehicles of smaller capacity since the
shipments in practice do not exceed 30 cubic meters.

5 CONCLUSIONS
In this work it has been presented a solution strategy
for long-term design of an efficient fleet. The approach
incorporates new factors that are critical to design and uses
control variables determined by an expert planner in order to
meet system requirements. In the future it is possible to
incorporate this strategy in a higher level of aggregation
within an organization to govern lowers decision levels such
as daily scheduling and routing of vehicles.

Table 3: Algorithm CK for periods in 2009


Table 4: Algorithm CK for complete horizon in 2009

Sindelen S.A ACK
A B C D E A B C D E
1 7 8 4 5 4 2 2 2 2

H
Y
93,582,920 69,987,580


90.00% 99.93%
Capacity %
35.19% 54.16%


6 ACKNOWLEDGMENTS
The authors are very grateful to DICYT (Scientific and
Technological Research Bureau), the Industrial Engineering
Department both of the University of Santiago of Chile.

7 REFERENCES
[1] Ballou, R.H., 2004, Logistics: Supply Chain
Management, Pearson.
[2] McGinnis,M.A., 1990, The Relative Importance of Cost
and Service in Freight Transportation Choice, Before
and After Deregulation, Transportation Journal, Vol. 30
(1), 12-19.
[3] Evers, P.F., Harper, D.V., Needham, P.M., 1996, The
determinants of Shipper Perceptions of Modes,
Transportation Journal, Vol. 36 (2),13-25.
[4] Lai, M.F., Lo, H.K., 2004, Ferry service network design:
optimal fleet size, routing, and scheduling.
Transportation Research, Part A 38, 305-328.
[5] Georgiadis, P., Vlachos, D., Iakovou, E., 2005, A
system dynamics modeling framework for the strategic
supply chain management of food chains. Journal of
Food Engineering, 70, 351364.



S

A B C D E M S
COST
NSET
1 Mx 6 3 4 2 5 20 5,183,370 100.00
2 Mx 6 3 4 2 5 20 5,493,470 100.00
3 Med 4 2 2 2 2 12 8,783,080 98.52
4 Mx 6 3 4 2 5 20 7,592,750 100.00
5 Mix 5 3 3 2 4 17 6,261,640 100.00
6 Med 4 2 2 2 2 12 5,564,170 98.52
7 Mix 5 3 3 2 4 17 6,591,570 100.00
8 Med 4 2 2 2 2 12 5,353,090 98.52
9 Mix 5 3 3 2 4 17 4,328,480 100.00
10 Med 4 2 2 2 2 12 6,094,590 98.52
11 Mx 6 3 4 2 5 20 4,961,310 100.00
12 Mx 6 3 4 2 5 20 3,333,080 100.00